OCR Filter Classifies Text Images to Reduce Computational Load
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Deployed OCR systems face high computational loads due to inefficient processing of images without text, leading to potential data loss and leakage of sensitive information, as they apply extensive pre-processing and post-processing pipelines even on non-textual image data.
Innovation Solution
A trained OCR filter, utilizing a convolutional neural network (CNN) or other image classification model, is employed to classify images as containing or not containing text, thereby filtering out non-textual images before full OCR processing, reducing unnecessary computations and securely storing sensitive data by using representations from internal layers.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If OCR applies extensive pre-processing and post-processing pipelines to all images, then text recognition accuracy is improved, but computational load and processing time increase significantly for images without text
Solution Approach 1:
The patent applies preliminary action by using a classification model to predict whether an image contains text before applying the full OCR pipeline. This preliminary classification step filters out images without text, preventing unnecessary pre-processing and post-processing computations while maintaining accuracy for images that do contain text.
Solution Approach 2:
The patent extracts the text detection function from the full OCR pipeline and separates it into a standalone classification model. This extracted component can independently evaluate images and return a binary classification (text present/absent), allowing the main OCR pipeline to be skipped for negative cases and thus reducing overall computational load.
2Reliability
If OCR processes all images through the complete pipeline, then comprehensive text detection is achieved, but sensitive client data may be exposed through unnecessary processing and storage
Solution Approach 1:
The classification model performs preliminary action by assessing image content before it enters the full OCR pipeline. Images classified as containing no text are stopped from processing, preventing sensitive data from being exposed through unnecessary analysis and storage operations, while maintaining complete detection for relevant images.
Solution Approach 2:
The patent extracts the classification decision-making logic from the full OCR system, creating a separate model that can independently determine whether images require processing. This extraction allows for selective processing that minimizes data exposure while maintaining detection reliability for images that do contain text.
3Quantity of substance
If the OCR system processes large volumes of image data, then comprehensive text recognition is achieved, but computational resources are wasted on images without text
Solution Approach 1:
The classification model performs preliminary triage on incoming image data, quickly identifying which images contain text and which do not. This preliminary action enables the system to process only the necessary subset of images through the full OCR pipeline, significantly reducing computational resource waste while maintaining comprehensive processing capability for relevant data.
Solution Approach 2:
The patent extracts the filtering function as a separate classification model that operates independently from the main OCR processing. This extracted filter component can efficiently screen large volumes of image data and route only relevant images to the full processing pipeline, minimizing energy consumption on unnecessary computations.
Data Source
AI summary
An OCR filter described herein filters non-textual files in scanned customer data from optical character recognition (OCR) and pattern analysis of text generated thereof for sensitive customer data. The OCR filter is trained on files labelled using feature values for features generated from OCR applied to the corresponding files. Moreover, the OCR filter stores internal representations of the files during training to avoid leaking potential sensitive customer data contained therein. Once trained, performance of the OCR filter in filtering files comprising image data without text is evaluated according to false positive rates and false negative rates by comparing classifications of the OCR filter to classifications according to feature values for features generated from OCR. Evaluation of the OCR filter ensures continued model performance and informs model updates.